Explainable AI Alert Visualization via Relevance Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current artificial intelligence systems, particularly in decision-making processes like misappropriation detection, lack explainability and interpretability, leading to regulatory concerns and limited user understanding of their decision-making processes.
Innovation Solution
A system that uses a machine learning model to generate a relevance visualization map by calculating feed-forward scoring and relevance visualization, allowing for the matching and display of interaction data features with known misappropriation patterns, and incorporates neural networks with backward propagation and layer-wise relevance propagation to enhance explainability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning models are used for misappropriation detection decisions, then detection accuracy and productivity are improved, but explainability and interpretability deteriorate
Solution Approach 1:
The patent introduces visualization maps as an intermediary between the machine learning model and the analyst. The visualization map translates the internal decision-making process of the AI model into a visual format that shows which features contributed to the misappropriation decision, allowing analysts to understand the reasoning without slowing down automated processing
Solution Approach 2:
The patent segments the decision-making process into visualizable components by creating a visualization map that breaks down the AI model's reasoning into individual feature contributions. This segmentation allows each feature's impact on the decision to be separately examined and understood
2Measurement precision
If complex machine learning models are deployed for pattern recognition, then detection precision is improved, but device complexity and ease of operation worsen
Solution Approach 1:
The patent creates a visual copy or representation of the model's internal state through the visualization map. This copy presents the complex decision-making process in a simplified visual format that is easier to understand and operate with, without requiring changes to the underlying complex model architecture
3Productivity
If AI systems make critical decisions autonomously, then productivity increases, but reliability and regulatory compliance deteriorate due to lack of explainability
Solution Approach 1:
The visualization map provides feedback to analysts about the AI model's reasoning process. This feedback loop allows analysts to verify and understand automated decisions in real-time, ensuring regulatory compliance and building trust while maintaining high automated processing volumes
Data Source
AI summary
A system for machine learning data pattern recognition for misappropriation identification is provided. The system comprises a controller configured for learning and identifying misappropriation data patterns. The controller is further configured to: receive interaction data associated with a received interaction, the interaction data comprising one or more features, wherein the one or more features are measurable characteristics of the interaction; calculate a feed-forward scoring of an input of the interaction data comprising one or more features; generate a relevance visualization map of the one or more features of the feed-forward scoring; match, using a machine learning model, the relevance visualization map of the received interaction to a visualization pattern associated with a known labeled misappropriation type, wherein the machine learning model is trained with known misappropriation data patterns; and display the relevance visualization map and the visualization pattern from the known misappropriation patterns with the known labeled misappropriation type.